Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77033.40052489293,
-      "required_pages_per_second_upper_estimate": 0.005329349390427169
+      "required_token_positions_per_second": 76920.53103505312,
+      "required_pages_per_second_upper_estimate": 0.005321540817226752
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9629.175065611616,
-      "required_pages_per_second_upper_estimate": 0.00403153451993064
+      "required_token_positions_per_second": 9615.06637938164,
+      "required_pages_per_second_upper_estimate": 0.00402562750762902
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19258.350131223233,
-      "required_pages_per_second_upper_estimate": 0.0040110064554876
+      "required_token_positions_per_second": 19230.13275876328,
+      "required_pages_per_second_upper_estimate": 0.004005129520946346
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28887.52519683485,
-      "required_pages_per_second_upper_estimate": 0.00173176302685325
+      "required_token_positions_per_second": 28845.19913814492,
+      "required_pages_per_second_upper_estimate": 0.0017292256442629382
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28887.52519683485,
-      "required_pages_per_second_upper_estimate": 0.0022458780212589787
+      "required_token_positions_per_second": 28845.19913814492,
+      "required_pages_per_second_upper_estimate": 0.0022425873563684936
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28887.52519683485,
-      "required_pages_per_second_upper_estimate": 0.37053815621701686
+      "required_token_positions_per_second": 28845.19913814492,
+      "required_pages_per_second_upper_estimate": 0.36999524298232345
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1028706387498346,
-    "producer_p95_seconds": 0.238108382000064,
-    "gpu_update_seconds": 0.9570913330792815,
-    "producer_batches_per_second": 9.72094673614154,
-    "gpu_batches_per_second": 1.0448323638901493,
-    "producer_headroom": 9.303833870486397,
+    "producer_batch_seconds": 0.2914910166247182,
+    "producer_p95_seconds": 0.2586335230007535,
+    "gpu_update_seconds": 0.9584957228961632,
+    "producer_batches_per_second": 3.4306374569596283,
+    "gpu_batches_per_second": 1.0433014734572092,
+    "producer_headroom": 3.288251329303174,
     "sustainable": true,
-    "buffer_seconds": 60.29675398399473,
+    "buffer_seconds": 60.38523054245829,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342028.147185368,
-    "upload_bytes_per_second": 5020124.577523501,
+    "generation_bytes_per_second": 356342.5158592566,
+    "upload_bytes_per_second": 5072144.418155734,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.39127837820797,
+    "estimated_full_shards_per_hour": 76.46280867417596,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.4798138580008526,
-      "bytes": 7393031,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.8912479129994608,
       "bytes": 1116344,
@@ -562,6 +556,12 @@
       "bytes": 2401890,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.469864372000302,
+      "bytes": 11763636,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -182,3 +182,4 @@
 {"step": 778200, "loss": 0.5231640338897705, "lr": 3e-05, "grad_norm": 0.24825772797739806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113906537906, "tokens_per_second": 152285.53860219131, "data_wait_s": 1.3607001164928079e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 778300, "loss": 0.4791816771030426, "lr": 3e-05, "grad_norm": 0.24315523523943586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113921334291, "tokens_per_second": 153813.61661321798, "data_wait_s": 1.4381999790202826e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 778400, "loss": 0.5778656005859375, "lr": 3e-05, "grad_norm": 0.2760404048411358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113935975517, "tokens_per_second": 151290.53802068773, "data_wait_s": 1.3567998394137248e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 778500, "loss": 0.5702077150344849, "lr": 3e-05, "grad_norm": 0.25043204536375985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113950515945, "tokens_per_second": 137848.96088460786, "data_wait_s": 1.4463999832514673e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd08a2ba7ce2a852a7cd0db07b999554b8cfe258c15f7eca4b5feaf5417e601b
-size 392226
+oid sha256:d31a55929d0446eb1e931b2c32939774de0eaf17dd9043aea7f2c5449cae492b
+size 394344